Rarbg, a popular torrent site, has announced its decision to shut down due to various reasons including COVID-19 complications, power price increase in data centers, inflation, and team members fighting in Europe.
The announcement states that the decision was unanimous, and the team can no longer bear the expenses of running the site.
Rarbg expressed regret and apologized for the decision to shut down.
Rarbg has been shut down, and the discussion on Hacker News speculates about illegal activities and how to avoid getting caught.
The thread highlights the potential of cryptocurrencies and blockchain technology in facilitating complex financial infrastructure and improving security in casual economic activity for torrent sharing.
Decentralized search engines and torrent sharing platforms are discussed, mainly the importance of the number of seeders for the viability of content. Different decentralized approaches to search exist, but there seems to be a lack of public search engines for this.
A group of AI technologists signed a statement that mitigating the risk of extinction from AI should be a global priority alongside other societal-scale risks such as pandemics and nuclear war.
While risks from AI exist, the focus should be on the people who control it rather than the technology itself, as the greatest risks historically come from those who control technology using it to accumulate power and wealth.
The urgent priority of AI risks should be mitigating the more likely risks posed by people negligently, recklessly or maliciously using AI systems, exacerbating inequality and undermining individual and collective freedom.
NVIDIA announced the DGX GH200, the first supercomputer with 100 terabytes of memory accessible to GPUs over NVLink, pairing NVIDIA Grace Hopper Superchip with the NVLink Switch System.
The DGX GH200 features a GPU complex baseboard interconnected with NVLink, with each GPU accessing other GPUs' memory at NVLink speed, and can form larger supercomputers.
The DGX GH200 is designed for the most demanding giant AI workloads, with potential speedups of 4x to 7x for deep learning recommendation models, terabyte-scale graph neural networks, or large data analytics workloads compared to the DGX A100.
Nvidia has announced the DGX SuperPOD, a scalable system for AI with 192 A100 GPUs and 15.36 petabytes of memory, improving AI development ease and affordability.
There is discussion around ad regulations, Google's TPU being better than Nvidia's AI hardware, and the future of tech giants and their diversification efforts.
The limitations of training an AI model with 100T parameters have been explored, along with ongoing research for larger multi-modal input training models and Nvidia's new Grace CPU.
Notes apps and bookmarking tools are marketed as a "second brain" that helps us remember everything.
The psychological concept of loss aversion causes people to hoard ideas, leading to mental overwhelm and an inability to remember important things.
Notes apps can be valuable in allowing us to forget and clear space for new ideas, but relying too heavily on them can lead to disappointment and the illusion of value.
The Japanese government will not enforce copyrights on data used in AI training, allowing AI to use any data for non-profit or commercial purposes, whether it is obtained legally or illegally.
Japan's government believes this no-copyright approach is necessary for the nation to remain competitive in the AI technology industry and plans to become a leader in AI technology with the help of ambitious plans and collaborations with local tech firms.
While some Japanese creators are concerned about the impact of AI on their work value, the academic and business sectors are pressing forward, and the effective implementation of AI could potentially boost Japan's GDP by 50% or more. Japan hopes to leverage Western literary resources for Japanese AI in exchange for the use of Japanese culture for Western AI training data.
Japan's government will not enforce copyrights on data used for AI training- There is debate about whether AI-generated content constitutes copyright infringement- The extraction of private information and concern for IP violations are potential issues with large language models (LLMs)
An anonymous feedback requested an article on how to become a systems engineer; it is an ambiguous term and involves complex management tasks beyond software development's scope.
The author proposes that becoming a systems engineer is about understanding the history and design of complex systems, developing hypotheses, refining models, and building up a library of experiences over time, which could be earned by slogging away at the job for years and years.
The author also acknowledges the existence of psychopathic individuals in the industry who make it their mission to wreck other people's world.
Reddit's API pricing would cost Apollo developer, Christian Selig, $20 million per year, after estimating that the 7 billion requests made for Apollo last month would cost $1.7 million per month.
Selig expressed disappointment at the high cost, given Reddit's earlier claims that pricing would be reasonable and based on reality.
If the API pricing is implemented, Apollo's free and Pro tiers would be unsustainable, and the Ultra subscription would have to cost at least $50 or $60 per year.